Semi-Markov Decision Process With Partial Information for Maintenance Decisions

A critical factor that prevents optimal scheduling of maintenance interventions is the uncertainty regarding the current condition of the asset under consideration, as well as the rate at which deterioration takes place. However, current maintenance modeling and optimization techniques assume that the condition of the asset is either known, or assumed to have an exponential deterioration rate. In this paper, we present a novel approach to maintenance modeling that removes such assumptions. Here, we employ a Partially Observable Semi-Markov Decision Process (POSMDP) for optimizing maintenance decisions, where the condition of the asset is not fully observable, and decision epochs occur at times following any other type of distribution. This method enables a more realistic way of modeling asset deterioration and optimizing maintenance schedules.